lfm3-vl:2b-mlx-bf16

7 9 minutes ago

LFM3-1.7B is a tri-hybrid model for local agents and specialized applications.

vision tools thinking 2b
... /
json
18a02e8b5963 ยท 2.9kB
{
"architectures": [
"Lfm3VlForConditionalGeneration"
],
"bos_token_id": 124894,
"downsample_factor": 2,
"dtype": "bfloat16",
"eos_token_id": 124900,
"image_token_id": 125030,
"model_type": "lfm3_vl",
"pad_token_id": 124893,
"projector_bias": true,
"projector_hidden_act": "gelu",
"projector_hidden_size": 4096,
"projector_layer_norm_eps": 1e-06,
"text_config": {
"bos_token_id": 124894,
"conv_L_cache": 7,
"conv_bias": false,
"dtype": "bfloat16",
"eos_token_id": 124900,
"head_dim": 128,
"hidden_size": 2048,
"initializer_range": 0.02,
"intermediate_size": 7680,
"layer_types": [
"conv",
"conv",
"full_attention",
"conv",
"sliding_attention",
"conv",
"full_attention",
"conv",
"sliding_attention",
"conv",
"full_attention",
"conv",
"sliding_attention",
"conv",
"full_attention",
"conv",
"sliding_attention",
"conv",
"full_attention",
"conv",
"sliding_attention",
"conv",
"full_attention",
"conv"
],
"max_position_embeddings": 128000,
"model_type": "lfm3",
"norm_eps": 1e-05,
"num_attention_heads": 16,
"num_hidden_layers": 24,
"num_key_value_heads": 4,
"pad_token_id": 124893,
"rope_parameters": {
"full_attention": {
"partial_rotary_factor": 0.5,
"rope_theta": 1000000.0,
"rope_type": "default"
},
"sliding_attention": {
"rope_theta": 10000,
"rope_type": "default"
}
},
"sliding_window": 256,
"tie_word_embeddings": true,
"use_cache": true,
"use_sandwich_norm": true,
"vocab_size": 128000
},
"tie_word_embeddings": true,
"transformers_version": "5.19.0.dev0",
"video_token_id": 125029,
"vision_config": {
"conv_kernel_size": 3,
"dtype": "bfloat16",
"hidden_size": 1024,
"in_channels": 3,
"initializer_range": 0.02,
"intermediate_size": 2560,
"layer_types": [
"conv",
"conv",
"full_attention",
"conv",
"full_attention",
"conv",
"full_attention",
"conv",
"full_attention",
"conv",
"full_attention",
"conv",
"full_attention",
"conv"
],
"model_type": "lfm3_vl_vision",
"norm_eps": 1e-05,
"num_attention_heads": 16,
"num_hidden_layers": 14,
"num_key_value_heads": 8,
"patch_size": 16,
"rope_parameters": {
"rope_theta": 1000000.0,
"rope_type": "axial"
},
"spatial_merge_size": 2,
"temporal_patch_size": 2
},
"vision_end_token_id": 125010,
"vision_start_token_id": 125009,
"auto_map": {
"AutoConfig": "configuration_lfm3_vl.Lfm3VlConfig",
"AutoModel": "modeling_lfm3_vl.Lfm3VlModel",
"AutoModelForImageTextToText": "modeling_lfm3_vl.Lfm3VlForConditionalGeneration"
}
}